Merge branch 'hv/config-from-blob'
[git] / sha1_name.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "tree.h"
5 #include "blob.h"
6 #include "tree-walk.h"
7 #include "refs.h"
8 #include "remote.h"
9
10 static int get_sha1_oneline(const char *, unsigned char *, struct commit_list *);
11
12 typedef int (*disambiguate_hint_fn)(const unsigned char *, void *);
13
14 struct disambiguate_state {
15         disambiguate_hint_fn fn;
16         void *cb_data;
17         unsigned char candidate[20];
18         unsigned candidate_exists:1;
19         unsigned candidate_checked:1;
20         unsigned candidate_ok:1;
21         unsigned disambiguate_fn_used:1;
22         unsigned ambiguous:1;
23         unsigned always_call_fn:1;
24 };
25
26 static void update_candidates(struct disambiguate_state *ds, const unsigned char *current)
27 {
28         if (ds->always_call_fn) {
29                 ds->ambiguous = ds->fn(current, ds->cb_data) ? 1 : 0;
30                 return;
31         }
32         if (!ds->candidate_exists) {
33                 /* this is the first candidate */
34                 hashcpy(ds->candidate, current);
35                 ds->candidate_exists = 1;
36                 return;
37         } else if (!hashcmp(ds->candidate, current)) {
38                 /* the same as what we already have seen */
39                 return;
40         }
41
42         if (!ds->fn) {
43                 /* cannot disambiguate between ds->candidate and current */
44                 ds->ambiguous = 1;
45                 return;
46         }
47
48         if (!ds->candidate_checked) {
49                 ds->candidate_ok = ds->fn(ds->candidate, ds->cb_data);
50                 ds->disambiguate_fn_used = 1;
51                 ds->candidate_checked = 1;
52         }
53
54         if (!ds->candidate_ok) {
55                 /* discard the candidate; we know it does not satisify fn */
56                 hashcpy(ds->candidate, current);
57                 ds->candidate_checked = 0;
58                 return;
59         }
60
61         /* if we reach this point, we know ds->candidate satisfies fn */
62         if (ds->fn(current, ds->cb_data)) {
63                 /*
64                  * if both current and candidate satisfy fn, we cannot
65                  * disambiguate.
66                  */
67                 ds->candidate_ok = 0;
68                 ds->ambiguous = 1;
69         }
70
71         /* otherwise, current can be discarded and candidate is still good */
72 }
73
74 static void find_short_object_filename(int len, const char *hex_pfx, struct disambiguate_state *ds)
75 {
76         struct alternate_object_database *alt;
77         char hex[40];
78         static struct alternate_object_database *fakeent;
79
80         if (!fakeent) {
81                 /*
82                  * Create a "fake" alternate object database that
83                  * points to our own object database, to make it
84                  * easier to get a temporary working space in
85                  * alt->name/alt->base while iterating over the
86                  * object databases including our own.
87                  */
88                 const char *objdir = get_object_directory();
89                 int objdir_len = strlen(objdir);
90                 int entlen = objdir_len + 43;
91                 fakeent = xmalloc(sizeof(*fakeent) + entlen);
92                 memcpy(fakeent->base, objdir, objdir_len);
93                 fakeent->name = fakeent->base + objdir_len + 1;
94                 fakeent->name[-1] = '/';
95         }
96         fakeent->next = alt_odb_list;
97
98         sprintf(hex, "%.2s", hex_pfx);
99         for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
100                 struct dirent *de;
101                 DIR *dir;
102                 sprintf(alt->name, "%.2s/", hex_pfx);
103                 dir = opendir(alt->base);
104                 if (!dir)
105                         continue;
106
107                 while (!ds->ambiguous && (de = readdir(dir)) != NULL) {
108                         unsigned char sha1[20];
109
110                         if (strlen(de->d_name) != 38)
111                                 continue;
112                         if (memcmp(de->d_name, hex_pfx + 2, len - 2))
113                                 continue;
114                         memcpy(hex + 2, de->d_name, 38);
115                         if (!get_sha1_hex(hex, sha1))
116                                 update_candidates(ds, sha1);
117                 }
118                 closedir(dir);
119         }
120 }
121
122 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
123 {
124         do {
125                 if (*a != *b)
126                         return 0;
127                 a++;
128                 b++;
129                 len -= 2;
130         } while (len > 1);
131         if (len)
132                 if ((*a ^ *b) & 0xf0)
133                         return 0;
134         return 1;
135 }
136
137 static void unique_in_pack(int len,
138                           const unsigned char *bin_pfx,
139                            struct packed_git *p,
140                            struct disambiguate_state *ds)
141 {
142         uint32_t num, last, i, first = 0;
143         const unsigned char *current = NULL;
144
145         open_pack_index(p);
146         num = p->num_objects;
147         last = num;
148         while (first < last) {
149                 uint32_t mid = (first + last) / 2;
150                 const unsigned char *current;
151                 int cmp;
152
153                 current = nth_packed_object_sha1(p, mid);
154                 cmp = hashcmp(bin_pfx, current);
155                 if (!cmp) {
156                         first = mid;
157                         break;
158                 }
159                 if (cmp > 0) {
160                         first = mid+1;
161                         continue;
162                 }
163                 last = mid;
164         }
165
166         /*
167          * At this point, "first" is the location of the lowest object
168          * with an object name that could match "bin_pfx".  See if we have
169          * 0, 1 or more objects that actually match(es).
170          */
171         for (i = first; i < num && !ds->ambiguous; i++) {
172                 current = nth_packed_object_sha1(p, i);
173                 if (!match_sha(len, bin_pfx, current))
174                         break;
175                 update_candidates(ds, current);
176         }
177 }
178
179 static void find_short_packed_object(int len, const unsigned char *bin_pfx,
180                                      struct disambiguate_state *ds)
181 {
182         struct packed_git *p;
183
184         prepare_packed_git();
185         for (p = packed_git; p && !ds->ambiguous; p = p->next)
186                 unique_in_pack(len, bin_pfx, p, ds);
187 }
188
189 #define SHORT_NAME_NOT_FOUND (-1)
190 #define SHORT_NAME_AMBIGUOUS (-2)
191
192 static int finish_object_disambiguation(struct disambiguate_state *ds,
193                                         unsigned char *sha1)
194 {
195         if (ds->ambiguous)
196                 return SHORT_NAME_AMBIGUOUS;
197
198         if (!ds->candidate_exists)
199                 return SHORT_NAME_NOT_FOUND;
200
201         if (!ds->candidate_checked)
202                 /*
203                  * If this is the only candidate, there is no point
204                  * calling the disambiguation hint callback.
205                  *
206                  * On the other hand, if the current candidate
207                  * replaced an earlier candidate that did _not_ pass
208                  * the disambiguation hint callback, then we do have
209                  * more than one objects that match the short name
210                  * given, so we should make sure this one matches;
211                  * otherwise, if we discovered this one and the one
212                  * that we previously discarded in the reverse order,
213                  * we would end up showing different results in the
214                  * same repository!
215                  */
216                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
217                                     ds->fn(ds->candidate, ds->cb_data));
218
219         if (!ds->candidate_ok)
220                 return SHORT_NAME_AMBIGUOUS;
221
222         hashcpy(sha1, ds->candidate);
223         return 0;
224 }
225
226 static int disambiguate_commit_only(const unsigned char *sha1, void *cb_data_unused)
227 {
228         int kind = sha1_object_info(sha1, NULL);
229         return kind == OBJ_COMMIT;
230 }
231
232 static int disambiguate_committish_only(const unsigned char *sha1, void *cb_data_unused)
233 {
234         struct object *obj;
235         int kind;
236
237         kind = sha1_object_info(sha1, NULL);
238         if (kind == OBJ_COMMIT)
239                 return 1;
240         if (kind != OBJ_TAG)
241                 return 0;
242
243         /* We need to do this the hard way... */
244         obj = deref_tag(parse_object(sha1), NULL, 0);
245         if (obj && obj->type == OBJ_COMMIT)
246                 return 1;
247         return 0;
248 }
249
250 static int disambiguate_tree_only(const unsigned char *sha1, void *cb_data_unused)
251 {
252         int kind = sha1_object_info(sha1, NULL);
253         return kind == OBJ_TREE;
254 }
255
256 static int disambiguate_treeish_only(const unsigned char *sha1, void *cb_data_unused)
257 {
258         struct object *obj;
259         int kind;
260
261         kind = sha1_object_info(sha1, NULL);
262         if (kind == OBJ_TREE || kind == OBJ_COMMIT)
263                 return 1;
264         if (kind != OBJ_TAG)
265                 return 0;
266
267         /* We need to do this the hard way... */
268         obj = deref_tag(lookup_object(sha1), NULL, 0);
269         if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
270                 return 1;
271         return 0;
272 }
273
274 static int disambiguate_blob_only(const unsigned char *sha1, void *cb_data_unused)
275 {
276         int kind = sha1_object_info(sha1, NULL);
277         return kind == OBJ_BLOB;
278 }
279
280 static int prepare_prefixes(const char *name, int len,
281                             unsigned char *bin_pfx,
282                             char *hex_pfx)
283 {
284         int i;
285
286         hashclr(bin_pfx);
287         memset(hex_pfx, 'x', 40);
288         for (i = 0; i < len ;i++) {
289                 unsigned char c = name[i];
290                 unsigned char val;
291                 if (c >= '0' && c <= '9')
292                         val = c - '0';
293                 else if (c >= 'a' && c <= 'f')
294                         val = c - 'a' + 10;
295                 else if (c >= 'A' && c <='F') {
296                         val = c - 'A' + 10;
297                         c -= 'A' - 'a';
298                 }
299                 else
300                         return -1;
301                 hex_pfx[i] = c;
302                 if (!(i & 1))
303                         val <<= 4;
304                 bin_pfx[i >> 1] |= val;
305         }
306         return 0;
307 }
308
309 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
310                           unsigned flags)
311 {
312         int status;
313         char hex_pfx[40];
314         unsigned char bin_pfx[20];
315         struct disambiguate_state ds;
316         int quietly = !!(flags & GET_SHA1_QUIETLY);
317
318         if (len < MINIMUM_ABBREV || len > 40)
319                 return -1;
320         if (prepare_prefixes(name, len, bin_pfx, hex_pfx) < 0)
321                 return -1;
322
323         prepare_alt_odb();
324
325         memset(&ds, 0, sizeof(ds));
326         if (flags & GET_SHA1_COMMIT)
327                 ds.fn = disambiguate_commit_only;
328         else if (flags & GET_SHA1_COMMITTISH)
329                 ds.fn = disambiguate_committish_only;
330         else if (flags & GET_SHA1_TREE)
331                 ds.fn = disambiguate_tree_only;
332         else if (flags & GET_SHA1_TREEISH)
333                 ds.fn = disambiguate_treeish_only;
334         else if (flags & GET_SHA1_BLOB)
335                 ds.fn = disambiguate_blob_only;
336
337         find_short_object_filename(len, hex_pfx, &ds);
338         find_short_packed_object(len, bin_pfx, &ds);
339         status = finish_object_disambiguation(&ds, sha1);
340
341         if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
342                 return error("short SHA1 %.*s is ambiguous.", len, hex_pfx);
343         return status;
344 }
345
346
347 int for_each_abbrev(const char *prefix, each_abbrev_fn fn, void *cb_data)
348 {
349         char hex_pfx[40];
350         unsigned char bin_pfx[20];
351         struct disambiguate_state ds;
352         int len = strlen(prefix);
353
354         if (len < MINIMUM_ABBREV || len > 40)
355                 return -1;
356         if (prepare_prefixes(prefix, len, bin_pfx, hex_pfx) < 0)
357                 return -1;
358
359         prepare_alt_odb();
360
361         memset(&ds, 0, sizeof(ds));
362         ds.always_call_fn = 1;
363         ds.cb_data = cb_data;
364         ds.fn = fn;
365
366         find_short_object_filename(len, hex_pfx, &ds);
367         find_short_packed_object(len, bin_pfx, &ds);
368         return ds.ambiguous;
369 }
370
371 const char *find_unique_abbrev(const unsigned char *sha1, int len)
372 {
373         int status, exists;
374         static char hex[41];
375
376         exists = has_sha1_file(sha1);
377         memcpy(hex, sha1_to_hex(sha1), 40);
378         if (len == 40 || !len)
379                 return hex;
380         while (len < 40) {
381                 unsigned char sha1_ret[20];
382                 status = get_short_sha1(hex, len, sha1_ret, GET_SHA1_QUIETLY);
383                 if (exists
384                     ? !status
385                     : status == SHORT_NAME_NOT_FOUND) {
386                         hex[len] = 0;
387                         return hex;
388                 }
389                 len++;
390         }
391         return hex;
392 }
393
394 static int ambiguous_path(const char *path, int len)
395 {
396         int slash = 1;
397         int cnt;
398
399         for (cnt = 0; cnt < len; cnt++) {
400                 switch (*path++) {
401                 case '\0':
402                         break;
403                 case '/':
404                         if (slash)
405                                 break;
406                         slash = 1;
407                         continue;
408                 case '.':
409                         continue;
410                 default:
411                         slash = 0;
412                         continue;
413                 }
414                 break;
415         }
416         return slash;
417 }
418
419 static inline int upstream_mark(const char *string, int len)
420 {
421         const char *suffix[] = { "@{upstream}", "@{u}" };
422         int i;
423
424         for (i = 0; i < ARRAY_SIZE(suffix); i++) {
425                 int suffix_len = strlen(suffix[i]);
426                 if (suffix_len <= len
427                     && !memcmp(string, suffix[i], suffix_len))
428                         return suffix_len;
429         }
430         return 0;
431 }
432
433 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags);
434 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf);
435
436 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
437 {
438         static const char *warn_msg = "refname '%.*s' is ambiguous.";
439         static const char *object_name_msg = N_(
440         "Git normally never creates a ref that ends with 40 hex characters\n"
441         "because it will be ignored when you just specify 40-hex. These refs\n"
442         "may be created by mistake. For example,\n"
443         "\n"
444         "  git checkout -b $br $(git rev-parse ...)\n"
445         "\n"
446         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
447         "examine these refs and maybe delete them. Turn this message off by\n"
448         "running \"git config advice.object_name_warning false\"");
449         unsigned char tmp_sha1[20];
450         char *real_ref = NULL;
451         int refs_found = 0;
452         int at, reflog_len, nth_prior = 0;
453
454         if (len == 40 && !get_sha1_hex(str, sha1)) {
455                 refs_found = dwim_ref(str, len, tmp_sha1, &real_ref);
456                 if (refs_found > 0 && warn_ambiguous_refs) {
457                         warning(warn_msg, len, str);
458                         if (advice_object_name_warning)
459                                 fprintf(stderr, "%s\n", _(object_name_msg));
460                 }
461                 free(real_ref);
462                 return 0;
463         }
464
465         /* basic@{time or number or -number} format to query ref-log */
466         reflog_len = at = 0;
467         if (len && str[len-1] == '}') {
468                 for (at = len-4; at >= 0; at--) {
469                         if (str[at] == '@' && str[at+1] == '{') {
470                                 if (str[at+2] == '-') {
471                                         if (at != 0)
472                                                 /* @{-N} not at start */
473                                                 return -1;
474                                         nth_prior = 1;
475                                         continue;
476                                 }
477                                 if (!upstream_mark(str + at, len - at)) {
478                                         reflog_len = (len-1) - (at+2);
479                                         len = at;
480                                 }
481                                 break;
482                         }
483                 }
484         }
485
486         /* Accept only unambiguous ref paths. */
487         if (len && ambiguous_path(str, len))
488                 return -1;
489
490         if (nth_prior) {
491                 struct strbuf buf = STRBUF_INIT;
492                 int detached;
493
494                 if (interpret_nth_prior_checkout(str, &buf) > 0) {
495                         detached = (buf.len == 40 && !get_sha1_hex(buf.buf, sha1));
496                         strbuf_release(&buf);
497                         if (detached)
498                                 return 0;
499                 }
500         }
501
502         if (!len && reflog_len)
503                 /* allow "@{...}" to mean the current branch reflog */
504                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
505         else if (reflog_len)
506                 refs_found = dwim_log(str, len, sha1, &real_ref);
507         else
508                 refs_found = dwim_ref(str, len, sha1, &real_ref);
509
510         if (!refs_found)
511                 return -1;
512
513         if (warn_ambiguous_refs &&
514             (refs_found > 1 ||
515              !get_short_sha1(str, len, tmp_sha1, GET_SHA1_QUIETLY)))
516                 warning(warn_msg, len, str);
517
518         if (reflog_len) {
519                 int nth, i;
520                 unsigned long at_time;
521                 unsigned long co_time;
522                 int co_tz, co_cnt;
523
524                 /* Is it asking for N-th entry, or approxidate? */
525                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
526                         char ch = str[at+2+i];
527                         if ('0' <= ch && ch <= '9')
528                                 nth = nth * 10 + ch - '0';
529                         else
530                                 nth = -1;
531                 }
532                 if (100000000 <= nth) {
533                         at_time = nth;
534                         nth = -1;
535                 } else if (0 <= nth)
536                         at_time = 0;
537                 else {
538                         int errors = 0;
539                         char *tmp = xstrndup(str + at + 2, reflog_len);
540                         at_time = approxidate_careful(tmp, &errors);
541                         free(tmp);
542                         if (errors)
543                                 return -1;
544                 }
545                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
546                                 &co_time, &co_tz, &co_cnt)) {
547                         if (!len) {
548                                 if (!prefixcmp(real_ref, "refs/heads/")) {
549                                         str = real_ref + 11;
550                                         len = strlen(real_ref + 11);
551                                 } else {
552                                         /* detached HEAD */
553                                         str = "HEAD";
554                                         len = 4;
555                                 }
556                         }
557                         if (at_time)
558                                 warning("Log for '%.*s' only goes "
559                                         "back to %s.", len, str,
560                                         show_date(co_time, co_tz, DATE_RFC2822));
561                         else {
562                                 die("Log for '%.*s' only has %d entries.",
563                                     len, str, co_cnt);
564                         }
565                 }
566         }
567
568         free(real_ref);
569         return 0;
570 }
571
572 static int get_parent(const char *name, int len,
573                       unsigned char *result, int idx)
574 {
575         unsigned char sha1[20];
576         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
577         struct commit *commit;
578         struct commit_list *p;
579
580         if (ret)
581                 return ret;
582         commit = lookup_commit_reference(sha1);
583         if (!commit)
584                 return -1;
585         if (parse_commit(commit))
586                 return -1;
587         if (!idx) {
588                 hashcpy(result, commit->object.sha1);
589                 return 0;
590         }
591         p = commit->parents;
592         while (p) {
593                 if (!--idx) {
594                         hashcpy(result, p->item->object.sha1);
595                         return 0;
596                 }
597                 p = p->next;
598         }
599         return -1;
600 }
601
602 static int get_nth_ancestor(const char *name, int len,
603                             unsigned char *result, int generation)
604 {
605         unsigned char sha1[20];
606         struct commit *commit;
607         int ret;
608
609         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
610         if (ret)
611                 return ret;
612         commit = lookup_commit_reference(sha1);
613         if (!commit)
614                 return -1;
615
616         while (generation--) {
617                 if (parse_commit(commit) || !commit->parents)
618                         return -1;
619                 commit = commit->parents->item;
620         }
621         hashcpy(result, commit->object.sha1);
622         return 0;
623 }
624
625 struct object *peel_to_type(const char *name, int namelen,
626                             struct object *o, enum object_type expected_type)
627 {
628         if (name && !namelen)
629                 namelen = strlen(name);
630         while (1) {
631                 if (!o || (!o->parsed && !parse_object(o->sha1)))
632                         return NULL;
633                 if (expected_type == OBJ_ANY || o->type == expected_type)
634                         return o;
635                 if (o->type == OBJ_TAG)
636                         o = ((struct tag*) o)->tagged;
637                 else if (o->type == OBJ_COMMIT)
638                         o = &(((struct commit *) o)->tree->object);
639                 else {
640                         if (name)
641                                 error("%.*s: expected %s type, but the object "
642                                       "dereferences to %s type",
643                                       namelen, name, typename(expected_type),
644                                       typename(o->type));
645                         return NULL;
646                 }
647         }
648 }
649
650 static int peel_onion(const char *name, int len, unsigned char *sha1)
651 {
652         unsigned char outer[20];
653         const char *sp;
654         unsigned int expected_type = 0;
655         unsigned lookup_flags = 0;
656         struct object *o;
657
658         /*
659          * "ref^{type}" dereferences ref repeatedly until you cannot
660          * dereference anymore, or you get an object of given type,
661          * whichever comes first.  "ref^{}" means just dereference
662          * tags until you get a non-tag.  "ref^0" is a shorthand for
663          * "ref^{commit}".  "commit^{tree}" could be used to find the
664          * top-level tree of the given commit.
665          */
666         if (len < 4 || name[len-1] != '}')
667                 return -1;
668
669         for (sp = name + len - 1; name <= sp; sp--) {
670                 int ch = *sp;
671                 if (ch == '{' && name < sp && sp[-1] == '^')
672                         break;
673         }
674         if (sp <= name)
675                 return -1;
676
677         sp++; /* beginning of type name, or closing brace for empty */
678         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
679                 expected_type = OBJ_COMMIT;
680         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
681                 expected_type = OBJ_TREE;
682         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
683                 expected_type = OBJ_BLOB;
684         else if (!prefixcmp(sp, "object}"))
685                 expected_type = OBJ_ANY;
686         else if (sp[0] == '}')
687                 expected_type = OBJ_NONE;
688         else if (sp[0] == '/')
689                 expected_type = OBJ_COMMIT;
690         else
691                 return -1;
692
693         if (expected_type == OBJ_COMMIT)
694                 lookup_flags = GET_SHA1_COMMITTISH;
695         else if (expected_type == OBJ_TREE)
696                 lookup_flags = GET_SHA1_TREEISH;
697
698         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
699                 return -1;
700
701         o = parse_object(outer);
702         if (!o)
703                 return -1;
704         if (!expected_type) {
705                 o = deref_tag(o, name, sp - name - 2);
706                 if (!o || (!o->parsed && !parse_object(o->sha1)))
707                         return -1;
708                 hashcpy(sha1, o->sha1);
709                 return 0;
710         }
711
712         /*
713          * At this point, the syntax look correct, so
714          * if we do not get the needed object, we should
715          * barf.
716          */
717         o = peel_to_type(name, len, o, expected_type);
718         if (!o)
719                 return -1;
720
721         hashcpy(sha1, o->sha1);
722         if (sp[0] == '/') {
723                 /* "$commit^{/foo}" */
724                 char *prefix;
725                 int ret;
726                 struct commit_list *list = NULL;
727
728                 /*
729                  * $commit^{/}. Some regex implementation may reject.
730                  * We don't need regex anyway. '' pattern always matches.
731                  */
732                 if (sp[1] == '}')
733                         return 0;
734
735                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
736                 commit_list_insert((struct commit *)o, &list);
737                 ret = get_sha1_oneline(prefix, sha1, list);
738                 free(prefix);
739                 return ret;
740         }
741         return 0;
742 }
743
744 static int get_describe_name(const char *name, int len, unsigned char *sha1)
745 {
746         const char *cp;
747         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
748
749         for (cp = name + len - 1; name + 2 <= cp; cp--) {
750                 char ch = *cp;
751                 if (hexval(ch) & ~0377) {
752                         /* We must be looking at g in "SOMETHING-g"
753                          * for it to be describe output.
754                          */
755                         if (ch == 'g' && cp[-1] == '-') {
756                                 cp++;
757                                 len -= cp - name;
758                                 return get_short_sha1(cp, len, sha1, flags);
759                         }
760                 }
761         }
762         return -1;
763 }
764
765 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
766 {
767         int ret, has_suffix;
768         const char *cp;
769
770         /*
771          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
772          */
773         has_suffix = 0;
774         for (cp = name + len - 1; name <= cp; cp--) {
775                 int ch = *cp;
776                 if ('0' <= ch && ch <= '9')
777                         continue;
778                 if (ch == '~' || ch == '^')
779                         has_suffix = ch;
780                 break;
781         }
782
783         if (has_suffix) {
784                 int num = 0;
785                 int len1 = cp - name;
786                 cp++;
787                 while (cp < name + len)
788                         num = num * 10 + *cp++ - '0';
789                 if (!num && len1 == len - 1)
790                         num = 1;
791                 if (has_suffix == '^')
792                         return get_parent(name, len1, sha1, num);
793                 /* else if (has_suffix == '~') -- goes without saying */
794                 return get_nth_ancestor(name, len1, sha1, num);
795         }
796
797         ret = peel_onion(name, len, sha1);
798         if (!ret)
799                 return 0;
800
801         ret = get_sha1_basic(name, len, sha1);
802         if (!ret)
803                 return 0;
804
805         /* It could be describe output that is "SOMETHING-gXXXX" */
806         ret = get_describe_name(name, len, sha1);
807         if (!ret)
808                 return 0;
809
810         return get_short_sha1(name, len, sha1, lookup_flags);
811 }
812
813 /*
814  * This interprets names like ':/Initial revision of "git"' by searching
815  * through history and returning the first commit whose message starts
816  * the given regular expression.
817  *
818  * For future extension, ':/!' is reserved. If you want to match a message
819  * beginning with a '!', you have to repeat the exclamation mark.
820  */
821 #define ONELINE_SEEN (1u<<20)
822
823 static int handle_one_ref(const char *path,
824                 const unsigned char *sha1, int flag, void *cb_data)
825 {
826         struct commit_list **list = cb_data;
827         struct object *object = parse_object(sha1);
828         if (!object)
829                 return 0;
830         if (object->type == OBJ_TAG) {
831                 object = deref_tag(object, path, strlen(path));
832                 if (!object)
833                         return 0;
834         }
835         if (object->type != OBJ_COMMIT)
836                 return 0;
837         commit_list_insert_by_date((struct commit *)object, list);
838         return 0;
839 }
840
841 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
842                             struct commit_list *list)
843 {
844         struct commit_list *backup = NULL, *l;
845         int found = 0;
846         regex_t regex;
847
848         if (prefix[0] == '!') {
849                 if (prefix[1] != '!')
850                         die ("Invalid search pattern: %s", prefix);
851                 prefix++;
852         }
853
854         if (regcomp(&regex, prefix, REG_EXTENDED))
855                 die("Invalid search pattern: %s", prefix);
856
857         for (l = list; l; l = l->next) {
858                 l->item->object.flags |= ONELINE_SEEN;
859                 commit_list_insert(l->item, &backup);
860         }
861         while (list) {
862                 char *p, *to_free = NULL;
863                 struct commit *commit;
864                 enum object_type type;
865                 unsigned long size;
866                 int matches;
867
868                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
869                 if (!parse_object(commit->object.sha1))
870                         continue;
871                 if (commit->buffer)
872                         p = commit->buffer;
873                 else {
874                         p = read_sha1_file(commit->object.sha1, &type, &size);
875                         if (!p)
876                                 continue;
877                         to_free = p;
878                 }
879
880                 p = strstr(p, "\n\n");
881                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
882                 free(to_free);
883
884                 if (matches) {
885                         hashcpy(sha1, commit->object.sha1);
886                         found = 1;
887                         break;
888                 }
889         }
890         regfree(&regex);
891         free_commit_list(list);
892         for (l = backup; l; l = l->next)
893                 clear_commit_marks(l->item, ONELINE_SEEN);
894         free_commit_list(backup);
895         return found ? 0 : -1;
896 }
897
898 struct grab_nth_branch_switch_cbdata {
899         int remaining;
900         struct strbuf buf;
901 };
902
903 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
904                                   const char *email, unsigned long timestamp, int tz,
905                                   const char *message, void *cb_data)
906 {
907         struct grab_nth_branch_switch_cbdata *cb = cb_data;
908         const char *match = NULL, *target = NULL;
909         size_t len;
910
911         if (!prefixcmp(message, "checkout: moving from ")) {
912                 match = message + strlen("checkout: moving from ");
913                 target = strstr(match, " to ");
914         }
915
916         if (!match || !target)
917                 return 0;
918         if (--(cb->remaining) == 0) {
919                 len = target - match;
920                 strbuf_reset(&cb->buf);
921                 strbuf_add(&cb->buf, match, len);
922                 return 1; /* we are done */
923         }
924         return 0;
925 }
926
927 /*
928  * Parse @{-N} syntax, return the number of characters parsed
929  * if successful; otherwise signal an error with negative value.
930  */
931 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
932 {
933         long nth;
934         int retval;
935         struct grab_nth_branch_switch_cbdata cb;
936         const char *brace;
937         char *num_end;
938
939         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
940                 return -1;
941         brace = strchr(name, '}');
942         if (!brace)
943                 return -1;
944         nth = strtol(name + 3, &num_end, 10);
945         if (num_end != brace)
946                 return -1;
947         if (nth <= 0)
948                 return -1;
949         cb.remaining = nth;
950         strbuf_init(&cb.buf, 20);
951
952         retval = 0;
953         if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
954                 strbuf_reset(buf);
955                 strbuf_add(buf, cb.buf.buf, cb.buf.len);
956                 retval = brace - name + 1;
957         }
958
959         strbuf_release(&cb.buf);
960         return retval;
961 }
962
963 int get_sha1_mb(const char *name, unsigned char *sha1)
964 {
965         struct commit *one, *two;
966         struct commit_list *mbs;
967         unsigned char sha1_tmp[20];
968         const char *dots;
969         int st;
970
971         dots = strstr(name, "...");
972         if (!dots)
973                 return get_sha1(name, sha1);
974         if (dots == name)
975                 st = get_sha1("HEAD", sha1_tmp);
976         else {
977                 struct strbuf sb;
978                 strbuf_init(&sb, dots - name);
979                 strbuf_add(&sb, name, dots - name);
980                 st = get_sha1_committish(sb.buf, sha1_tmp);
981                 strbuf_release(&sb);
982         }
983         if (st)
984                 return st;
985         one = lookup_commit_reference_gently(sha1_tmp, 0);
986         if (!one)
987                 return -1;
988
989         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
990                 return -1;
991         two = lookup_commit_reference_gently(sha1_tmp, 0);
992         if (!two)
993                 return -1;
994         mbs = get_merge_bases(one, two, 1);
995         if (!mbs || mbs->next)
996                 st = -1;
997         else {
998                 st = 0;
999                 hashcpy(sha1, mbs->item->object.sha1);
1000         }
1001         free_commit_list(mbs);
1002         return st;
1003 }
1004
1005 /* parse @something syntax, when 'something' is not {.*} */
1006 static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1007 {
1008         if (len || name[1] == '{')
1009                 return -1;
1010
1011         strbuf_reset(buf);
1012         strbuf_add(buf, "HEAD", 4);
1013         return 1;
1014 }
1015
1016 static int reinterpret(const char *name, int namelen, int len, struct strbuf *buf)
1017 {
1018         /* we have extra data, which might need further processing */
1019         struct strbuf tmp = STRBUF_INIT;
1020         int used = buf->len;
1021         int ret;
1022
1023         strbuf_add(buf, name + len, namelen - len);
1024         ret = interpret_branch_name(buf->buf, &tmp);
1025         /* that data was not interpreted, remove our cruft */
1026         if (ret < 0) {
1027                 strbuf_setlen(buf, used);
1028                 return len;
1029         }
1030         strbuf_reset(buf);
1031         strbuf_addbuf(buf, &tmp);
1032         strbuf_release(&tmp);
1033         /* tweak for size of {-N} versus expanded ref name */
1034         return ret - used + len;
1035 }
1036
1037 /*
1038  * This reads short-hand syntax that not only evaluates to a commit
1039  * object name, but also can act as if the end user spelled the name
1040  * of the branch from the command line.
1041  *
1042  * - "@{-N}" finds the name of the Nth previous branch we were on, and
1043  *   places the name of the branch in the given buf and returns the
1044  *   number of characters parsed if successful.
1045  *
1046  * - "<branch>@{upstream}" finds the name of the other ref that
1047  *   <branch> is configured to merge with (missing <branch> defaults
1048  *   to the current branch), and places the name of the branch in the
1049  *   given buf and returns the number of characters parsed if
1050  *   successful.
1051  *
1052  * If the input is not of the accepted format, it returns a negative
1053  * number to signal an error.
1054  *
1055  * If the input was ok but there are not N branch switches in the
1056  * reflog, it returns 0.
1057  */
1058 int interpret_branch_name(const char *name, struct strbuf *buf)
1059 {
1060         char *cp;
1061         struct branch *upstream;
1062         int namelen = strlen(name);
1063         int len = interpret_nth_prior_checkout(name, buf);
1064         int tmp_len;
1065
1066         if (!len) {
1067                 return len; /* syntax Ok, not enough switches */
1068         } else if (len > 0) {
1069                 if (len == namelen)
1070                         return len; /* consumed all */
1071                 else
1072                         return reinterpret(name, namelen, len, buf);
1073         }
1074
1075         cp = strchr(name, '@');
1076         if (!cp)
1077                 return -1;
1078
1079         len = interpret_empty_at(name, namelen, cp - name, buf);
1080         if (len > 0)
1081                 return reinterpret(name, namelen, len, buf);
1082
1083         tmp_len = upstream_mark(cp, namelen - (cp - name));
1084         if (!tmp_len)
1085                 return -1;
1086
1087         len = cp + tmp_len - name;
1088         cp = xstrndup(name, cp - name);
1089         upstream = branch_get(*cp ? cp : NULL);
1090         /*
1091          * Upstream can be NULL only if cp refers to HEAD and HEAD
1092          * points to something different than a branch.
1093          */
1094         if (!upstream)
1095                 die(_("HEAD does not point to a branch"));
1096         if (!upstream->merge || !upstream->merge[0]->dst) {
1097                 if (!ref_exists(upstream->refname))
1098                         die(_("No such branch: '%s'"), cp);
1099                 if (!upstream->merge) {
1100                         die(_("No upstream configured for branch '%s'"),
1101                                 upstream->name);
1102                 }
1103                 die(
1104                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1105                         upstream->merge[0]->src);
1106         }
1107         free(cp);
1108         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1109         strbuf_reset(buf);
1110         strbuf_addstr(buf, cp);
1111         free(cp);
1112         return len;
1113 }
1114
1115 int strbuf_branchname(struct strbuf *sb, const char *name)
1116 {
1117         int len = strlen(name);
1118         int used = interpret_branch_name(name, sb);
1119
1120         if (used == len)
1121                 return 0;
1122         if (used < 0)
1123                 used = 0;
1124         strbuf_add(sb, name + used, len - used);
1125         return len;
1126 }
1127
1128 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1129 {
1130         strbuf_branchname(sb, name);
1131         if (name[0] == '-')
1132                 return -1;
1133         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1134         return check_refname_format(sb->buf, 0);
1135 }
1136
1137 /*
1138  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1139  * notably "xyz^" for "parent of xyz"
1140  */
1141 int get_sha1(const char *name, unsigned char *sha1)
1142 {
1143         struct object_context unused;
1144         return get_sha1_with_context(name, 0, sha1, &unused);
1145 }
1146
1147 /*
1148  * Many callers know that the user meant to name a committish by
1149  * syntactical positions where the object name appears.  Calling this
1150  * function allows the machinery to disambiguate shorter-than-unique
1151  * abbreviated object names between committish and others.
1152  *
1153  * Note that this does NOT error out when the named object is not a
1154  * committish. It is merely to give a hint to the disambiguation
1155  * machinery.
1156  */
1157 int get_sha1_committish(const char *name, unsigned char *sha1)
1158 {
1159         struct object_context unused;
1160         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1161                                      sha1, &unused);
1162 }
1163
1164 int get_sha1_treeish(const char *name, unsigned char *sha1)
1165 {
1166         struct object_context unused;
1167         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1168                                      sha1, &unused);
1169 }
1170
1171 int get_sha1_commit(const char *name, unsigned char *sha1)
1172 {
1173         struct object_context unused;
1174         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1175                                      sha1, &unused);
1176 }
1177
1178 int get_sha1_tree(const char *name, unsigned char *sha1)
1179 {
1180         struct object_context unused;
1181         return get_sha1_with_context(name, GET_SHA1_TREE,
1182                                      sha1, &unused);
1183 }
1184
1185 int get_sha1_blob(const char *name, unsigned char *sha1)
1186 {
1187         struct object_context unused;
1188         return get_sha1_with_context(name, GET_SHA1_BLOB,
1189                                      sha1, &unused);
1190 }
1191
1192 /* Must be called only when object_name:filename doesn't exist. */
1193 static void diagnose_invalid_sha1_path(const char *prefix,
1194                                        const char *filename,
1195                                        const unsigned char *tree_sha1,
1196                                        const char *object_name,
1197                                        int object_name_len)
1198 {
1199         struct stat st;
1200         unsigned char sha1[20];
1201         unsigned mode;
1202
1203         if (!prefix)
1204                 prefix = "";
1205
1206         if (!lstat(filename, &st))
1207                 die("Path '%s' exists on disk, but not in '%.*s'.",
1208                     filename, object_name_len, object_name);
1209         if (errno == ENOENT || errno == ENOTDIR) {
1210                 char *fullname = xmalloc(strlen(filename)
1211                                              + strlen(prefix) + 1);
1212                 strcpy(fullname, prefix);
1213                 strcat(fullname, filename);
1214
1215                 if (!get_tree_entry(tree_sha1, fullname,
1216                                     sha1, &mode)) {
1217                         die("Path '%s' exists, but not '%s'.\n"
1218                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1219                             fullname,
1220                             filename,
1221                             object_name_len, object_name,
1222                             fullname,
1223                             object_name_len, object_name,
1224                             filename);
1225                 }
1226                 die("Path '%s' does not exist in '%.*s'",
1227                     filename, object_name_len, object_name);
1228         }
1229 }
1230
1231 /* Must be called only when :stage:filename doesn't exist. */
1232 static void diagnose_invalid_index_path(int stage,
1233                                         const char *prefix,
1234                                         const char *filename)
1235 {
1236         struct stat st;
1237         const struct cache_entry *ce;
1238         int pos;
1239         unsigned namelen = strlen(filename);
1240         unsigned fullnamelen;
1241         char *fullname;
1242
1243         if (!prefix)
1244                 prefix = "";
1245
1246         /* Wrong stage number? */
1247         pos = cache_name_pos(filename, namelen);
1248         if (pos < 0)
1249                 pos = -pos - 1;
1250         if (pos < active_nr) {
1251                 ce = active_cache[pos];
1252                 if (ce_namelen(ce) == namelen &&
1253                     !memcmp(ce->name, filename, namelen))
1254                         die("Path '%s' is in the index, but not at stage %d.\n"
1255                             "Did you mean ':%d:%s'?",
1256                             filename, stage,
1257                             ce_stage(ce), filename);
1258         }
1259
1260         /* Confusion between relative and absolute filenames? */
1261         fullnamelen = namelen + strlen(prefix);
1262         fullname = xmalloc(fullnamelen + 1);
1263         strcpy(fullname, prefix);
1264         strcat(fullname, filename);
1265         pos = cache_name_pos(fullname, fullnamelen);
1266         if (pos < 0)
1267                 pos = -pos - 1;
1268         if (pos < active_nr) {
1269                 ce = active_cache[pos];
1270                 if (ce_namelen(ce) == fullnamelen &&
1271                     !memcmp(ce->name, fullname, fullnamelen))
1272                         die("Path '%s' is in the index, but not '%s'.\n"
1273                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1274                             fullname, filename,
1275                             ce_stage(ce), fullname,
1276                             ce_stage(ce), filename);
1277         }
1278
1279         if (!lstat(filename, &st))
1280                 die("Path '%s' exists on disk, but not in the index.", filename);
1281         if (errno == ENOENT || errno == ENOTDIR)
1282                 die("Path '%s' does not exist (neither on disk nor in the index).",
1283                     filename);
1284
1285         free(fullname);
1286 }
1287
1288
1289 static char *resolve_relative_path(const char *rel)
1290 {
1291         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1292                 return NULL;
1293
1294         if (!startup_info)
1295                 die("BUG: startup_info struct is not initialized.");
1296
1297         if (!is_inside_work_tree())
1298                 die("relative path syntax can't be used outside working tree.");
1299
1300         /* die() inside prefix_path() if resolved path is outside worktree */
1301         return prefix_path(startup_info->prefix,
1302                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1303                            rel);
1304 }
1305
1306 static int get_sha1_with_context_1(const char *name,
1307                                    unsigned flags,
1308                                    const char *prefix,
1309                                    unsigned char *sha1,
1310                                    struct object_context *oc)
1311 {
1312         int ret, bracket_depth;
1313         int namelen = strlen(name);
1314         const char *cp;
1315         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1316
1317         memset(oc, 0, sizeof(*oc));
1318         oc->mode = S_IFINVALID;
1319         ret = get_sha1_1(name, namelen, sha1, flags);
1320         if (!ret)
1321                 return ret;
1322         /*
1323          * sha1:path --> object name of path in ent sha1
1324          * :path -> object name of absolute path in index
1325          * :./path -> object name of path relative to cwd in index
1326          * :[0-3]:path -> object name of path in index at stage
1327          * :/foo -> recent commit matching foo
1328          */
1329         if (name[0] == ':') {
1330                 int stage = 0;
1331                 const struct cache_entry *ce;
1332                 char *new_path = NULL;
1333                 int pos;
1334                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1335                         struct commit_list *list = NULL;
1336                         for_each_ref(handle_one_ref, &list);
1337                         return get_sha1_oneline(name + 2, sha1, list);
1338                 }
1339                 if (namelen < 3 ||
1340                     name[2] != ':' ||
1341                     name[1] < '0' || '3' < name[1])
1342                         cp = name + 1;
1343                 else {
1344                         stage = name[1] - '0';
1345                         cp = name + 3;
1346                 }
1347                 new_path = resolve_relative_path(cp);
1348                 if (!new_path) {
1349                         namelen = namelen - (cp - name);
1350                 } else {
1351                         cp = new_path;
1352                         namelen = strlen(cp);
1353                 }
1354
1355                 strncpy(oc->path, cp,
1356                         sizeof(oc->path));
1357                 oc->path[sizeof(oc->path)-1] = '\0';
1358
1359                 if (!active_cache)
1360                         read_cache();
1361                 pos = cache_name_pos(cp, namelen);
1362                 if (pos < 0)
1363                         pos = -pos - 1;
1364                 while (pos < active_nr) {
1365                         ce = active_cache[pos];
1366                         if (ce_namelen(ce) != namelen ||
1367                             memcmp(ce->name, cp, namelen))
1368                                 break;
1369                         if (ce_stage(ce) == stage) {
1370                                 hashcpy(sha1, ce->sha1);
1371                                 oc->mode = ce->ce_mode;
1372                                 free(new_path);
1373                                 return 0;
1374                         }
1375                         pos++;
1376                 }
1377                 if (only_to_die && name[1] && name[1] != '/')
1378                         diagnose_invalid_index_path(stage, prefix, cp);
1379                 free(new_path);
1380                 return -1;
1381         }
1382         for (cp = name, bracket_depth = 0; *cp; cp++) {
1383                 if (*cp == '{')
1384                         bracket_depth++;
1385                 else if (bracket_depth && *cp == '}')
1386                         bracket_depth--;
1387                 else if (!bracket_depth && *cp == ':')
1388                         break;
1389         }
1390         if (*cp == ':') {
1391                 unsigned char tree_sha1[20];
1392                 int len = cp - name;
1393                 if (!get_sha1_1(name, len, tree_sha1, GET_SHA1_TREEISH)) {
1394                         const char *filename = cp+1;
1395                         char *new_filename = NULL;
1396
1397                         new_filename = resolve_relative_path(filename);
1398                         if (new_filename)
1399                                 filename = new_filename;
1400                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1401                         if (ret && only_to_die) {
1402                                 diagnose_invalid_sha1_path(prefix, filename,
1403                                                            tree_sha1,
1404                                                            name, len);
1405                         }
1406                         hashcpy(oc->tree, tree_sha1);
1407                         strncpy(oc->path, filename,
1408                                 sizeof(oc->path));
1409                         oc->path[sizeof(oc->path)-1] = '\0';
1410
1411                         free(new_filename);
1412                         return ret;
1413                 } else {
1414                         if (only_to_die)
1415                                 die("Invalid object name '%.*s'.", len, name);
1416                 }
1417         }
1418         return ret;
1419 }
1420
1421 /*
1422  * Call this function when you know "name" given by the end user must
1423  * name an object but it doesn't; the function _may_ die with a better
1424  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1425  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1426  * you have a chance to diagnose the error further.
1427  */
1428 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1429 {
1430         struct object_context oc;
1431         unsigned char sha1[20];
1432         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1433 }
1434
1435 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1436 {
1437         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1438 }